Claude Skills

Shared Claude Code skills for planning, code review, and documentation.

Workflow: From Idea to Implementation

/brainstorm [topic]
    │
    ├── Phase 1: Understand — gather context, ask questions one at a time
    ├── Phase 2: Explore — propose 2-3 approaches with trade-offs
    ├── Phase 3: Design — present design in sections, validate each
    ├── Phase 4: Plan — enter plan mode, review via plannotator
    │
    └── Phase 5: Execute
            ├── Start now — implement task by task
            └── Ralphex plan — autonomous execution via ralphex CLI

Start any feature, refactor, or investigation with /brainstorm. It walks through collaborative discovery, produces a validated design, then enters plan mode where plannotator opens a visual UI for annotation and approval. After approval, either implement manually or hand off to ralphex for autonomous execution.

Skills

brainstorm

The main planning skill. Conversational dialogue that turns ideas into validated designs with structured plan output.

/brainstorm add webhook support to our API

ralph-review

Automatic pre-PR iterative review pipeline modeled on umputun's 4-stage flow. Every phase passes findings to a fresh fixer subagent (verify → baseline → fix → validate → commit) so the orchestrator never reads code or runs tests itself. Severity tagging (CRITICAL/MAJOR/MINOR) drives codex's minor-only early-exit; a /tmp/ralphex-progress-<slug>.txt scratchpad (where <slug> is the branch name with / and other unsafe chars replaced) threads context across phases.

1. Review phase 1 — 5 parallel agents (quality, implementation, testing, simplification, documentation); iters 2+ narrow to 2 critical-only agents; loop ≤5.
2. Review phase 2 — code smells single pass.
3. Review phase 3 — codex external review (background, severity-based exit, ≤10 iterations).
4. Review phase 4 — critical-only safety net (single pass).

/ralph-review           # diff against main
/ralph-review develop   # diff against develop

Requires the codex CLI on PATH for review phase 3 (skipped gracefully if missing).

github-pr

Copilot-aware GitHub PR review loop. Polls for Copilot reviews (with required re-request between rounds), resolves resolved threads, and iterates until the PR is clean. Wraps gh CLI + GraphQL.

/github-pr 42

multi-review

Three-reviewer adversarial code review against a base branch. Runs Codex (via direct codex exec), a fresh Opus subagent (via Task), and Gemini (via the gemini CLI) in parallel, merges findings, and presents them as a single plannotator pass before applying any fix in one commit.

/multi-review

Prerequisites:

• thinking-tools plugin installed (provides the model-invocable ask-codex Skill).
• plannotator binary on PATH — the skill invokes plannotator annotate via Bash (the bundled /plannotator-annotate slash command has disable-model-invocation: true and is user-only by design).
• gemini CLI on PATH (auth via GEMINI_API_KEY env or gcloud auth application-default login).

ask-gemini

Thin wrapper around the local gemini CLI for second-opinion runs on diffs or quick adversarial reviews. Best used standalone when you want a focused Gemini pass outside multi-review.

ask gemini about this diff

md-to-pdf

Convert Markdown files to clean A4 PDFs with Mermaid diagram support via pre-render and SVG inlining.

/md-to-pdf docs/architecture.md

docx

Read, edit, and export Microsoft Word documents. Supports text replacement, table cell editing, and PDF export.

/docx read contract.docx
/docx edit contract.docx '{"old text": "new text"}'

notion-annotate

Remote-friendly alternative to plannotator. Pushes a markdown file to Notion so the user can read and comment on it from any device, then pulls comments back as git-review style feedback.

/notion-annotate push docs/plan.md
/notion-annotate pull docs/plan.md

skill-vetting

Security vetting for third-party skills. Downloads to /tmp, runs automated scanner, manual code review, utility assessment.

/skill-vetting https://github.com/someone/some-skill

skill-writer

Create and improve Claude Code skills following official best practices and Agent Skills spec.

/skill-writer create a new review skill

Companion Plugins

These plugins extend the workflow with additional capabilities:

Plugin	What it adds
backnotprop/plannotator	Visual plan review and annotation UI in browser
umputun/ralphex	Autonomous plan execution + ralphex-plan skill for plan creation
umputun/cc-thingz	ask-codex (GPT-5 second opinion), dialectic (opposing agents), root-cause-investigator (5-Why)
umputun/revdiff	TUI for reviewing diffs, files, and documents with inline annotations — outputs structured results for AI agents

/plugin marketplace add backnotprop/plannotator
/plugin install plannotator@plannotator --scope user

/plugin marketplace add umputun/ralphex
/plugin install ralphex --scope user

/plugin marketplace add umputun/cc-thingz
/plugin install thinking-tools --scope user

/plugin marketplace add umputun/revdiff
/plugin install revdiff@umputun-revdiff --scope user
brew install umputun/apps/revdiff

Installation

/plugin marketplace add fitz123/claude-skills
/plugin install claude-skills --scope user

Or test locally:

claude --plugin-dir /path/to/claude-skills

License

MIT